The Choice Between Flexible Exchange Rates, Capital Control and the Currency Board in Asian Countries: A Perspective from the “Impossible Trinity”

THE CHOICE BETWEEN FLEXIBLE EXCHANGE

RATES, CAPITAL CONTROL AND THE CURRENCY

BOARD IN ASIAN COUNTRIES: A PERSPECTIVE

FROM THE `̀ IMPOSSIBLE TRINITY''�

By KOICHI HAMADA{ and YOSUKE TAKEDA{{Economic and Social Research Institute, Cabinet Of®ce of Japan

{Department of Economics, Sophia University

We attempt to compare adjustment costs under exchange rate regimes in East Asianeconomies during their recovery processes. The criteria are the degree of overshooting inexchange rates, the changes in country risks, and the severity and duration of the recoveryprocesses. Linear ranking is dif®cult. Managed rates with capital control worked formacroeconomic performance despite the welfare loss due to blocking capital ¯ows. Thecurrency board system worked well for stability, but recent experiences of Argentina andHong Kong were de¯ationary. Under ¯exible rates, many economies that received IMF grantssuffered a drastic initial downturn but later recovered vigorously.JEL Classi®cation Numbers: F31, F32, F33.

1. Introduction: exchange rate regimes and adjustment costs

A number of different approaches have been used in attempting to explain the causes of therecent East Asian crises.1 Before it proved possible to establish such causes, most countrieshad started to recover from the severe recession. In this paper, rather than focusing on thecauses of the crises, we study the recovery processes and the sensitivity of recoveries to theexchange rate regimes that were adopted in those countries at the outset of the crisis. Ourregional focus is on the cases of Asian economies that have suffered from recent currencycrises, although Latin American experiences are often referred to for comparative purposes.

Unfortunately, since the data observation period is still extremely short, it is dif®cult toexplain comprehensively the relationship between the recovery process and the monetaryregime adopted. Nevertheless, we have attempted to provide country information that maybe useful in assessing the strength and weakness of alternative exchange rate regimes as ameans of coping with currency crises.2

We start with the traditional proposition of `̀ the impossible trinity''. A country cannothave all three goalsÐexchange rate stability, monetary independence, and capital

� We are indebted to Levis Kochin and Kar-yiu Wong for valuable comments at the Asian Crisis Conference,and to Roger Downey for helpful suggestions concerning data availability. Yosuke Takeda is grateful to theZengin Foundation for their ®nancial support.

{ Correspondence: Department of Economics, Sophia University. 7-1 Kioi-cho, Chiyoda-ku, Tokyo 102-8554,Japan. Phone: �81-3-3238-3209, Fax: �81-3-3238-3086, Email: [email protected]

1 The comprehensive literature on the East Asian crises are Radelet and Sachs (1998a, b) and Furman andStiglitz (1998).

2 Frankel (1999a, b) have the same motivation as ours. Our analysis differs from theirs, however, in focusing onthe East Asian crises and in quantifying the cost of adjustment during the recovery process.

± 429 ±# Japanese Economic Association 2001.

The Japanese Economic ReviewVol. 52, No. 4, December 2001

Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK.

mobilityÐsimultaneously under any exchange rate regime, i.e. ¯exible exchange rate(`¯oat'), currency board or capital control. It can attain a pair of objectivesÐthe ®rst twounder capital control, the last two under pure ¯oat, or the ®rst and third under a currencyboardÐbut not all three. In terms of what cannot be attained, each exchange rate regime hasto incur the adjustment costs on the outset of ®nancial crisis. The East Asian crisis was noexception. From this standpoint, we quantify the adjustment costs associated with theadoption of each exchange rate regime during the recovery process.

The ®ve economies under our consideration are Thailand, Indonesia, Malaysia, SouthKorea and Hong Kong, with occasional references to four Latin American countries:Mexico, Chile, Argentina and Brazil. These are classi®ed roughly in Table 1, with theapproximate starting month of the currency crisis. Many Asian economies under con-sideration (except Hong Kong) had adopted ®xed exchange rates or a crawling peg to themarket basket, with a heavy emphasis on the dollar. After the crisis, Indonesia, Korea andThailand adopted a ¯exible exchange rate system, presumably on the advice of the IMF.Thailand ¯oated its currency the baht in July 1997, though it had been pegging it to acurrency basket dominated by the US dollar. One month after the adoption of the ¯oat inThailand, Indonesia followed by ¯oating its currency, the rupiah. The IMF objected to theidea of adopting the currency board system, which was popular in Indonesian politicalcircles. Hong Kong has long been under the currency board system. In September 1998,Malaysia took a control on capital out¯ow, which essentially prohibits the repatriation offoreign capital with a maturity of less than a year. For a reference, this can be contrasted withthe control on capital in¯ow practised in Chile until recently, where capital in¯ows werediscouraged by the zero-interest deposit requirement for a part of the in¯ow. The Chileanmeasure was taken in order to avoid the incidence of sudden out¯ows of foreign capital.Chile started imposing control over its capital in¯ow from 1991, where its exchange ratesystem could be called `̀ the exchange rate within crawling bands''. As a result of thediscouraging effect of this capital control on the in¯ow of capital, we understand that suchrestrictions have now been at least temporarily suspended.

In Latin America, Brazil and Mexico also adopted a ¯exible exchange rate regime on therecommendation of the IMF. (The IMF has not always recommended the ¯exible exchangerate system, however: for example, it advised Bulgaria and Estonia to opt for the currencyboard system; and Argentina adopted a currency board arrangement.)

Table 1

Exchange Rate Regimes during the Currency Crisisa

Exchange rate regime East Asia Latin America

Flexible exchange rate Indonesia (rupiah, Aug. 1997)Korea (won, Nov. 1997)Thailand (baht, Jul. 1997)

Brazil (real, Dec. 1994)Mexico ( peso, Dec. 1994)

Currency board Hong Kong (dollar, Oct. 1997) Argentina ( peso, Dec. 1994)

Capital control Malaysia (ringgit, Aug. 1997) Chileb ( peso)

aCurrency and starting month of currency crisis in parentheses.bThere seems to be no record of a crisis in Chile, but we include the country among our sample in order to describehow it installed the control on capital in¯ow and to compare Chile's performance with those experiences that are notbased on capital control.


The Japanese Economic Review

2. Quantifying the adjustment costs

For our sample economies, we attempted to evaluate the costs of adjustment under eachexchange rate regime, according to four criteria. The ®rst is the degree of overshooting innominal and real exchange rates implied by our portfolio approach to exchange rates,explained below. The second is a change in country risk measured by risk premium,observed as the deviation from the exact interest rate parity condition under certainty. Thisrisk preference can also be understood in the framework of the portfolio approach. The thirdis the strength of the monetary contraction policy that the crisis necessitated. This cost ariseswhatever regime is adopted. The fourth and ®nal indicator is the severity and the duration ofthe recovery processes after a currency crisis.

By comparing these measures for the countries under consideration, we can assess howthe above costs of adjustment varied in the recovery from a ®nancial crisis under alternativeexchange rate regimes. We have ®rst to distinguish the strength of the initial shock, and therecovery process corresponds to the movement of the dynamic process after the shock. Wepresent a simple portfolio model of the exchange rate determination, similar to Kouri (1976)and Branson and Henderson (1985), taking the currency of Indonesia, the rupiah, as theexample.

Suppose that Indonesians hold only the asset denominated in rupiah, but that residents inthe rest of the world hold the assets denominated in rupiah and those denominated in dollars.Let us denote the exchange rate of the rupiah in terms of the dollar as e. Note that this is thereverse of the usual exchange rate. Let the total asset that the rest of the world possesses be Zrupiah. Then the balance of payments of Indonesia is a function of the exchange rate e, andthe amount of indebtedness Z. The balance of payments is a decreasing function of theexchange rate e and an increasing function of the indebtedness Z. In terms of the increase inZ, that is, the negative of the balance of payments of Indonesia, one obtains

dZ=dt � f (Z, e), (1)

where f Z , 0, and f e . 0.The portfolio balance equation expresses the relationship that people in the rest of the

world hold a higher proportion of the Indonesian asset in their portfolio if the expected rateof appreciation of the value of the Indonesian currency is higher. That is, denoting theexpectation by operator E,

(eZ)=(W � eZ) � g(ð), (2)

where ð � E[(de=dt)=e], and g9( ) . 0. If we impose the assumption of rationalexpectations such that E[(de=dt)=e] � (de=dt)=e, we obtain from (2) above, the following:

de=dt � h(Z, e), (3)

where hz . 0, and he . 0.Strictly speaking, the portfolio balance is meaningful for the nominal exchange rate e, and

the current account balance is meaningful for the real exchange rate because the currentaccount is considered to respond to the real exchange rate. For the moment this aspect is nottaken into consideration, though further development of the portfolio approach shouldcertainly do so.

In Figure 1, the phase diagram of the simultaneous equation system (1) and (3) is drawn asCC and PP. CC indicates the combination of e and Z that keeps the current account ofIndonesia in balance, or that maintains the value of Indonesian asset held by the rest of the


K. Hamada and Y. Takeda: Exchange Rate Regimes in Asian Countries

world constant. This is an intrinsically stable relationship, and the value of Z increases in theleft side of CC and decreases in the right. PP indicates the combination of e and Z that keepthe portfolio balance of the rest of the world. This is an intrinsically unstable relationship, sothat e increases above PP and decreases below PP. The combination of these two balancescreates a phase diagram around the intersection of CC and PPÐpoint AÐof the well-knownsaddlepoint type. Under changes in exogenous factors, exchange rate e jumps to the saddlestable path and the balance of payment adjusts gradually to the new equilibrium.

Before the crisis, the prospect of the Indonesian economy was so bright that Indonesianswere willing to invest even more of their savings, or to borrow from abroad. At that time thisperception was shared by the lenders as well. The rest of the world too was willing to hold alarge amount of Indonesian debt. Indonesia's future appeared bright, and the country riskwas considered small. Thus the portfolio balance PP was located to the right in the ®gure.Equilibrium was at a point like A, where Indonesian debt was large and the value ofIndonesian rupiah was high. Then, all of a sudden, the asset demand for the asset in rupiahdeclined precipitously, and the new equilibrium shifted to a point like B. Since Z can moveonly slowly, only e jumps, and the path of variables takes the trajectory like A through B9 toB. As will be shown below, a similar situation occurred in many countries, includingThailand and South Korea.

The model thus predicts ®rst the sudden overshooting depreciation in an Asian currencyby the dislocation of demand for the currency, and then the process by which the currentaccount of the Asian country gradually improves. The prediction of this model, surprisingly,applies equally well to the experiences of Asian countries, and even to those of some LatinAmerican countries. Figure 2 shows the changes in exchange rates after the dislocation ofcurrency demand and the following slow adjustment in current accounts. In most countries(except Brazil and Chile) shown, one can detect jumps in exchange rates and the reversal ofthe current account from de®cit to surplus. In our context we can interpret this as follows.When market participants suddenly realize that they have been over-optimistic about acountry's future income, then the ¯ow relation, equation (1) (CC in Figure 1), shifts to theleft because participants no longer regard the returns on their investments as reasonable.Moreover, the stock relation, equation (3) (PP in Figure 1), shifts drastically downward,indicating the precipitating fall of the exchange rate.

This model analyses a ¯oating exchange rate regime where the exchange rate is

e

A

B

CC

PP

Z

B′

P′P′

Figure 1. Portfolio approach to the exchange rate



determined freely. This is the regime that the IMF often prefers to recommend to ailingnations. The IMF might have implicitly endorsed the ®xed exchange rate practice before theonset of a crisis, but after the crisis in Asia it recommended the ¯exible exchange rateregime, with the ®scal and monetary austerity that may help the national economy shift CCand PP respectively and sustain its exchange rate. The economy should then return to a new,less extravagant, equilibrium position. Residents living in any economy that is distinct fromthe classical, money-neutral economy would suffer from this sudden change in the exchangerate and the following adjustment process.

As we have explained, one cannot sustain the impossible trilogy of a ®xed exchange rate,free capital mobility and autonomy of monetary policy. The ¯exible exchange rate regime

Figure 2A. Latin America

25000

24000

23000

22000

21000

0

1000

Q2 Q3 Q4 Q11994

Q2 Q3 Q4 Q11995

Q2 Q3 Q4 Q11996

Q2 Q3 Q4 Q11997

Q2 Q3 Q4 Q11998

Q2

Quarter

US

$m0.9982

0.9984

0.9986

0.9988

0.999

0.9992

0.9994

0.9996

0.9998

1

1.0002

per

US

$

Current account Nominal exchange rate

(a) Argentina

212000

210000

28000

26000

24000

22000

0

2000

4000

Q2 Q3 Q4 Q11994

Q2 Q3 Q4 Q11995

Q2 Q3 Q4 Q11996

Q2 Q3 Q4 Q11997

Q2 Q3 Q4 Q11998

Q2 Q3 Q4 Q11999

Q2 Q3 Q4

Quarter

US

$m

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2


per

US

$

(b) Brazil



arises when the ®rst of the three is given up. However, this not the only possible regime. Thecapital control regime and the currency board regime are two other feasible options.

The capital control changes the slope of the portfolio relationship and also the speed ofadjustment. The proportionate Tobin tax by rate ô on all the capital transactions willsubstitute g((1ÿ ô)ð) for g(ð). It is easy to see that the arrows of movements in the phasediagram become steeper. In this portfolio asset model, capital transaction taxation willincrease rather than decrease the volatility of exchange rates. On the other hand, if controland a deterrent to capital out¯ow are adopted, as in Malaysia, then the portfolio relation PPwill be shifted to the right and the equilibrium exchange rate for the local currency willinitially appreciate. Whether this increase is offset by the reluctance of potential investors to

Figure 2A. (continued )

22500

22000

21500

21000

2500

0

500

Q3 Q11990

Q3 Q11991

Q3 Q11992

Q3 Q11993

Q3 Q11994

Q3 Q11995

Q3 Q11996

Q3 Q11997

Q3 Q11998

Q3 Q11999

Q3

Quarter

US

$m

0

100

200

300

400

500

600

per

US

$


(c) Chile

29000

28000

27000

26000

25000

24000

23000

22000

21000

0

1000

Q2 Q3 Q4 Q11994

Q2 Q3 Q4 Q11995

Q2 Q3 Q4 Q11996

Q2

Quarter

US

$m

0

1

2

3

4

5

6

7

8


per

US

$

(d) Mexico

1993



invest in the country because of the unexpected imposition of control remains to be seen. Incase of the Chilean type of control, this effect of unexpected control does not exist, butgeneral discouragement against capital in¯ow remains. Naturally, the control tends to reducethe value of the home currency.

The ®xed exchange rate, including the currency board system, makes exchange rate apolicy variable. It changes the nature of the differential equations. The exchange rate is nolonger an endogenous, forward-looking variable, but a policy variable to be ®xed by themonetary authorities. Accordingly, it is no longer a jumping variable, either. Instead, moneysupply is no longer a policy variable. The ¯ow equation CC and the stock relation PP areforced to intersect for a determined value of the exchange rate. For this to occur, domestic

Figure 2B. East Asia

23000

22500

22000

21500

21000

2500

0

500

1000

1500

Q4 Q11996

Q2 Q3 Q4 Q11997

Q2 Q3 Q4 Q11998

Q2 Q3 Q4

Quarter

US

$m0

2000

4000

6000

8000

10000

12000

14000

per

US

$


(a) Indonesia

210000

25000

0

5000

10000

15000

Q2 Q3 Q4 Q11997

Q2 Q3 Q4 Q11998

Q2 Q3 Q4 Q11999

Q2

Quarter

US$

m

0

200

400

600

800

1000

1200

1400

1600

1800

per

US$


(b) South Korea

1996



price levels and the domestic interest rate will vary a great deal. Thus, the main cost to thecountry that adopts the currency board system is the high interest rate that comes fromthe speculative attack. A speculative attack could occur as long as the ®xed parity of thecurrency is less than perfectly credible. It draws the reserves from the system rapidly enoughto make the domestic interest rate extremely high. In Hong Kong, and Argentina, this iscoped with technical devices designed to broaden the money base. The cost of ®xing theexchange rate by the currency board system is thus the burden of high rates of interest for thecountry.

The adjustment mechanism differs, therefore, depending on the regime. So does the costof adjustment to the crisis countries. In other words, adjustment costs to employment, pricestability and income distribution are hidden behind equations (1) and (2). Once thedislocation of the asset demand has occurred, some adjustment costs are inevitable. We haveto cultivate an open-minded view on the relative costs and bene®ts of the various adjustmentmechanisms. Rather than arguing for or against the IMF scheme, we have to comparepossible alternatives, such as the currency board system, the ¯oating exchange rate and thecapital controls of Chile or Malaysia, against the criterion of how these different systemsaffect the adjustment costs that inevitably arise after the dislocation of capital demand. Tosum up, the strength of the initial shock corresponds to a downward shift of the PP curve inFigure 1, and the income loss in the adjustment process is the propagation cost (see Hamada,2000).

As already mentioned, in order to assess the relative merits of the various regimes, we haveto distinguish between the strength of the initial shocks on one hand and the cost of theensuing adjustment on the other (Table 2). The former may be represented by the initialexchange rate movements and by the increase in country risks. The latter can be representedby the degree of monetary contraction that was necessary to prevent the currency fromdevaluing, by the output loss during the adjustment period, by the duration of the resultingrecession and by the duration of the continuing high country risk. We will compare theseindexes across the regimes. Statistical ®gures are mostly drawn from International

26000

25000

24000

23000

22000

21000

0

1000

2000

3000

4000

5000

Q11996

Q2 Q3 Q4 Q11997

Q2 Q3 Q4 Q11998

Q2 Q3 Q4 Q11999

Quarter

US

$m

0

5

10

15

20

25

30

35

40

45

50

per

US

$


(c) Thailand

Figure 2B. (continued )



Financial Statistics,3 occasionally supplemented by recent statistics issued by the centralbanks in each country.

Because our interest is in the severity and duration of the adjustment process in the crisiscountries, we need sample periods that are long enough both to detect the start impulse of a®nancial crisis and to trace its eruption into a full-blown shock. In East Asia, the crisis beganin July 1997, when Thailand requested the IMF assistance. From Thailand it then spread toIndonesia, Malaysia, South Korea and Hong Kong. Therefore, our sample period for EastAsia starts at the beginning of 1997 and extends to the present.

Similarly, in December 1994 after the `̀ peso problem'' arose in Mexico, the peso wasplaced under a ¯exible exchange rate. This seemed to have a contagious effect on other LatinAmerican countries, especially Brazil and Argentina. For the sake of a comparison with EastAsia, our sample period for Latin America is also three years, from the beginning of 1994.We describe below the measures taken in chronological order, from Latin America to EastAsia.

2.1 Degree of overshooting

First, we observe the movements of exchange rates, both nominal and real. Strictly speaking,the nominal exchange rate is a determinant of the portfolio balance equation in our portfolioapproach, while the real exchange rate determines the balance of payments equation. In theevent of a ®nancial crisis, both exchange rates `̀ overshoot'', as is indicated by the dynamictrajectory of exchange rates following a downward shift of the portfolio balance curve inFigure 1. The instantaneous magnitude of a crisis can be measured by the degree of volatilityof both rates when the ®nancial crisis took hold. By de®nition, the volatility is greater under¯exible exchange rates, so that this is not a fair measure for the ¯exible system in ourcomparison of the performance of alternative systems. However, we consider the move-ments of both nominal and real exchange rates as measures representing the initial shocksunder ¯exible rates.

Table 2

Indexes of Impulse and Propagation

Indicator Variable

Impulse h Overshooting of Exchange Rates j Nominal Exchange Ratej Real Exchange Rate

h Country Risk j Interest Rates Differentialj Depreciation-Adjusted Risk Premium

Propagation h Country Risk j Interest Rates Differentialj Depreciation-Adjusted Risk Premium

h Monetary Contraction j Growth Rate of M2h Production Recovery j Real Value of the IIP de¯ated by the CPI.

3 We describe these data in the Appendix.



Nominal rate

To normalize the monthly exchange rates, we divided by the exchange rate level on January1994 for Latin America (Figures 3 and 5), or by that on January 1997 for East Asia (Figures 4and 6). Figure 3 shows how rapidly the nominal rate of Brazilian real devaluated ahead of theMexican crisisÐmore than six times in six months. At the time of the Mexican crisis,however, the real was not so responsive. Though the devaluation of Mexican peso was lessdramatic compared with the real, its value was halved in terms of the US dollar in just over a

0

1

2

3

4

5

6

7

8

JAN MAR MAY JUL1994

SEP NOV JAN MAR MAY JUL1995

SEP NOV JAN MAR MAY JUL1996

SEP NOV

Jan.

199

4=1

Argentina Brazil Chile Mexico

Figure 3. Nominal exchange rates in Latin America, 1994±1996

Figure 4. Nominal exchange rates in East Asia, 1997±1999



quarter after December 1994. In contrast, the pesos in both Argentina and Chile kept theirexchange rates vis-aÁ-vis the US dollar steady.

In Figure 4 we can see a similarly steep devaluation in the nominal exchange rate of theIndonesian rupiah after November 1997. The devaluation was much more drastic than othercurrencies during the Asian ®nancial crisis. In comparison, the Thai baht and the Koreanwon were devaluated substantially to less than half their former values. For the Indonesianrupiah, there were two troughs in the rate (or two peaks in the dollar exchange rate), one at thebeginning of 1998 and the other in June 1998, like `̀ a triple jump''. The ®rst took place at theonset of the crisis, while the second was triggered by the domestic political turmoil later.Since the end of 1998, however, the nominal rate has been steady at around three or fourtimes (in terms of the US dollar) the rate before the crisis. In other words, the rupiah has keptits value, after devaluation, at a third or a quarter of its value relative to the dollar before theonset of crisis.

The real exchange rate

In real terms, it was Mexico's peso, and not the Brazilian real, that was hardest hit by thecurrency devaluation among the four Latin American countries (Figure 5). The real ex-change rate of the Brazilian real was appreciating, rather than depreciating, by about 20%owing to the country's four-digit hyperin¯ation (i.e. real appreciation rather than nominaldepreciation). In this sense, Brazil is in a unique situation relative to the other three LatinAmerican countries considered.4

The real exchange rate of the rupiah depreciated by approximately the same degree as thenominal rate. In Figure 6 we can see the triple jump of the real exchange rate more clearlythan the nominal rate during the same periods.

0.5

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2

JAN1994

MAR MAY JUL SEP NOV JAN1995


MAR MAY JUL SEP NOV

Jan.

199

4=1


Figure 5. Real exchange rates in Latin America, 1994±1996

4 For differences in the dynamic responses of nominal and real exchange rates, see Branson and Henderson(1985, pp. 775±7).



This indicates some of the relevance of exchange rate determination models with short-run ®xed prices as in Dornbusch (1976). Price levels of many countries are rather rigid, andthe real exchange rate often moves in the same direction as the nominal rate. This indicatesthat the world is more like the Dornbusch model, or the portfolio balance model discussedbelow. Exchange rates did overshoot. Though the Dornbusch model is criticized for its lackof microeconomic foundations, it seems to explain the dramatic path of exchange rates betterthan the non-jumping model of Obstfeld and Rogoff (1995), which has a relativelysophisticated micro foundation. Incidentally, our diagrammatic exposition, which takesboth the real and the nominal exchange rate on the same axis in our portfolio approach, canbe justi®ed from these observations. One should however study more carefully a strikinganomaly of Brazil, that the real exchange rate moves in a different direction from the nominalrate.

2.2 Country risk

Second, we observed changes in `̀ country risk'' for international investors, which affect thecountries concerned as well as foreign investors. In the absence of capital control, anincrease in country risks implies that investors require higher rate of return to invest in thecountry, primarily because of the perceived risks. This is an indication of how investorsregard the economic security of the country in question.

When a country is under capital control, the country risk can be seen also as an indicatorof the shadow price of the control. In the short run, the capital control aims both to preventrapid currency devaluation and to sustain the autonomy of monetary policy. However, in thelong run, the dif®culty in sustaining free capital mobility leads international investors torequire a higher risk premium for assets denominated by the currency of the controlledcountry when they invest money in the country's assets. Therefore, the increase in currencyrisk indicators may be interpreted also as the long-run side-effects of capital control.

For an exposition, suppose that a US investor plans to invest either in an asset denominated

Figure 6. Real exchange rates in East Asia, 1997±1999



by the Malaysian ringgit or in a domestic one denominated by the US dollar. The interestrates are i in Malyasia and i� in the USA, respectively. We assume that Malaysia imposes acontrol on capital out¯ow, which functions like a proportionate `̀ Tobin tax'' by a rate ô on allthe capital transactions. Assuming that the investor requires a risk premium ä for the ringgitasset, we have the following de®nition of the country risk:

i� � ä � iÿ _e

eÿ ô,

where the exchange rate is in terms of the ringgit to the US dollar. ô is not the actual tax ratebut the tax rate equivalent of the capital control. The no-arbitrage condition is also derivedfrom the equilibrium of the general portfolio balance by countries, and can be called `̀ aninterest rate parity condition with country risk adjustment''.

By de®nition, the risk premium should be measured as ä, that is, as the residual of theinterest rates differential iÿ i�minus an expected depreciation rate of the ringgit _e=e minusthe tax equivalent of capital control, ô. Our ideal measure of country risk should encompassthe degree of capital mobility as well as the risk premium. The degree of capital mobility isrepresented by a sum of the risk premium and the tax rate. In the world of certainty, we do nothave any dif®culty in measuring ä because the actual proportionate change in exchange rateequals _e=e. However, under uncertainty the expected change in the exchange rate is notequal to its actual change. Moreover, the existence of the `̀ forward discount bias'' tells usthat the interest parity hypothesis cannot be empirically maintained whether it is uncoveredor covered. Intuitively, two interpretations of the bias are given: a time-varying risk premiumon foreign exchange, and expectation error by investors. Our risk premium measure wouldexpress the time-varying risk premium, unless any expectation errors existed. And our storyis simple. Instead, if expectation errors in market participants are signi®cantly large andnegative in a critical situation, then the risk premium calculated by substituting the expectedchange in exchange rates by the actual change will ¯uctuate considerably, and will suddenlybecome negative when the currency is unexpectedly devaluated. In any case, the ex postdepreciation rate would be an insuf®cient substitute for the ex ante one that is needed incalculating the interest rate parity under the uncertainty about the future exchange rate.

In addition, we cannot observe the tax equivalent ô, either. Nevertheless, Edwards (1999)estimates the tax equivalent of capital controls in Chile, assuming the capital in¯ow staysthere for 180 days, one year and three years, respectively. The estimate is given as

ô � ëi�1ÿ ë

rk

� �,

where i� is an international interest rate that captures the opportunity cost of the reserverequirement, ë is the proportion of the funds that has to be deposited at the central bank, r isthe period of time (months) that the deposit has to be kept in the central bank, and k is thematurity of the funds. We can similarly estimate the tax equivalent of capital control inMalaysia. We have to rely on the tax data, however, again on an ex post basis. While theChilean capital control was exercised ex ante, the Malaysian control did not allow the fundsto be repatriated once they were invested there under the belief that there would be no such atax. Consequently, even if we estimate the ex post tax rate in Malaysia, it may not be theappropriate measure of ôÐsame problem as for the depreciation rate.

In spite of many quali®cations, we calculate two measures of the country risk. The ®rst isthe `̀ devaluation-adjusted risk premium'', de®ned as a residual of an interest rate differentialiÿ i�minus an ex post depreciation rate _e=e of the exchange rate of the ringgit. The second



is the interest rate differential itself, iÿ i�. The latter is free from the problem of theexpectation errors, but it does not take account of the expected depreciation. In that case, therisk premium measure entails the peso problem even while the currency is actuallydepreciating. Without the expectation errors, the former would be more desirable for ourtime-varying country risk.5 Thus, though both indicators are far from perfect in measuringour country risk, we make use of information that these two measures provide. Consideringthe continuity of our data series, in the actual calculation we use a deposit rate in eachcountry and the US three-month LIBOR.

Devaluation-adjusted risk premium

Let us start with the devaluation-adjusted risk premium. Before the Mexican crisis started,Brazil had faced dif®culties in raising funds from the international ®nancial market.However, throughout the period of the Mexican crisis, the devaluation-adjusted country riskrapidly decreased from 84% in the third quarter of 1994 to 11% in the fourth quarter of 1996(Figure 7). In contrast to the improvement in the Brazilian country risk just after atremendous decrease in the country risk at the onset of the crisis, there was an increase in theMexican country risk. The peak rate 37% in the second quarter of 1995 was comparable tothe country risk of 41% for Brazil at that time.

On the other hand, though the country risk of Chile had been more stable than the abovetwo countries under a ¯exible rate system, it was on average higher (9.4%) than that ofArgentina (3.7%) under the currency board system during most of the period. Surprisingly,

5 Obstfeld and Taylor (1997) analyse the interest rate parity during the Great Depression, when the forwardexchange markets were absent. They ®nd that `̀ the Great Depression, perhaps as part of a much broaderinterwar phase of disintegration, stands out as an event that transformed the world capital market and leftinterest arbitrage differentials higher and more volatile than ever before.'' The non-existence of a well-organized forward market is common in most of the present Asian ®nancial markets.

2300

2250

2200

2150

2100

250

0

50

100

Q11994

Q2 Q3 Q4 Q11995

Q2 Q3 Q4 Q11996

Q2 Q3 Q4

%


Figure 7. Devaluation-adjusted risk premium, Latin America, 1994±1996



the risk of Chile was 33% in the ®rst quarter of 1995, greater than the 28% for Brazil.Because Chile had been under capital control since 1991, the increase in country risktriggered by the contagious Mexican crisis must have been re¯ected in an increase in the riskpremium. The risk premium was evaluated by international investors who faced theconstraint and the uncertainty generated by the Chilean capital control. However, the riskpremium after the crisis became lower than before, down to an average rate of 6.6%.

Like the case of Mexico, in Asian countries we ®nd an enormous decrease in the measureof devaluation-adjusted country risk at the time of each crisis, probably because of the errorsin expecting the rapid currency devaluation (Figure 8). In Thailand, the risk premium was64% in the ®rst quarter of 1998. Similarly, South Korea recorded the highest: 63%throughout the period. In both countries the miscalculation of investors concerning therecovery of the domestic currency produced this ex post risk premium; however, the premiain both countries were reduced in the following quarter. Conversely, in Indonesia, partlybecause of the political turmoil after the crisis, the ex post risk was very large (181%) in thethird quarter of 1998 in spite of the moderate rise in the value of the rupiah. The recovery ofthe baht also brought about a rebound in the risk premium (50%) for Thailand. At present,the Indonesian premium appears to be decreasing.

In the aftershock of the crisis of Indonesia, Malaysia imposed a control on capital out¯owsin September 1998. The effect of the capital control can be seen in the risk premium for thethird quarter of 1998: the risk premium then was 28%, while in the next quarter it wasreduced to 0.6%. From both the Chilean and Malaysian cases, the effects of capital controlon the country risk seem to have only a short, though it was certainly effective, life.

Interest rates differential

In order to determine the possibility of expectation errors in the index of devaluation-adjusted risk premia, we compare the conventional interest rate differential with thedepreciation-adjusted risk premium measure. In the presence of the very large expectation

Figure 8. Devaluation-adjusted risk premium, East Asia, 1997±1999



errors at the onset of the crisis, as seems often to be the case with ®nancial crisis, the ex postdepreciation-adjusted premium suddenly becomes negative. The depreciation-adjusted riskpremium must expose a downward and deep protuberance at the start of the crisis, in contrastwith the sluggish movement of the conventional risk premium measured by the interest ratesdifferential. If otherwiseÐthat is, if we observe similar ¯uctuations between both thedepreciation-adjusted premium and the interest rate differentialÐthen both indicators maywell be analysed as indictors of the time-varying country risk.

Certainly, according to Figure 9, the Brazilian country risk rapidly improved throughoutthe Mexican crisis, and thereafter moved parallel to the Mexican risk for a while. However,because there are no signs of any changes in the Mexican interest rate differential, theobserved drop in the above risk premium measure at the onset of the Mexican crisis can bejudged to have been caused by expectation errors.

The relatively stable but high country risk of Chile is also shown in the interest ratedifferential. The risk rate was on average higher than that of Argentina. It is likely to re¯ectthe effect of Chilean capital control on the sentiment of international investors who wereanxious about the tax-equivalent costs.

Similarly, in Figure 10 we ®nd a very large decrease in the Indonesian devaluation-adjusted risk premium measure, owing to the expectation errors. As well as in Indonesia, theexpectation errors seem to occur in Korea, Thailand and Malaysia before the crisis started.The high country risks in Korea and Thailand decreased quarter by quarter. The Indonesianspike in country risk probably re¯ects concern about the political turmoil following thecrisis.

2.3 Degree of monetary contraction policy

Third, we focus on the degree of monetary contraction that was necessitated by the currencycrisis. Let us consider ®rst the countries under the currency board system, that is, Argentinaand Hong Kong. They faced a credibility problem for international investors concerning

0

10

20

30

40

50

60

70

Q11994

Q2 Q3 Q4 Q11995

Q2 Q3 Q4 Q11996

Q2 Q3 Q4

%


Figure 9. Interest rate differentials, Latin America, 1994±1996



whether the government would maintain the ®xed currency value. Once such credibility islost, vigorous speculative attacks are likely to take place against the currency. In order tocope with such speculative attacks in the event of a crisis, the government must continue tomaintain higher interest rates, or to contract the money supply to the level less than the levelthat is desirable for the macroeconomy. Consequently, this curtailed money supply duringthe crisis could be a cause of the adjustment costs, which would have been forgone under¯exible rates.

However, this consideration applies to any country, not just those restricted to the currencyboard system. Whether a country is under the currency board system or a ®xed exchangerate, monetary contraction is necessitated by the system when reserves are depleted. This isthe advantage of the system that enforces the monetary discipline, but at the same time thecontraction may be harmful for the real economy. On the other hand, the ¯exible rate systemleaves monetary policy autonomous, and allows the economy to counteract recession. The¯exible exchange rate smoothes the wave of the recession if autonomy is used withdiscretion. The Indonesian case seems to indicate that the undisciplined monetary expansionat the outset of the crisis left its burden on the economy for many years to come. In thissituation, the IMF reinforces economies such as Indonesia with a high interest rate policy.The suppressing effects of such interest rates on real output must not be overlooked.

As an indicator of the growth rate of the money supply, we measured the quarterlypercentage change in M2. In Figure 11 we can generally see the successive growth in moneysupply for most Latin American countries during the crisis. However, two obvious ex-ceptions were Brazil and Argentina. In particular, a decrease in money supply (ÿ13%) in the®rst quarter of 1995 is conspicuous in Argentina under the currency board system. Underthis system the monetary authority, at the onset of the Mexican crisis, probably had tocontract the money supply because of the out¯ow of reserves. On the other hand, Brazil,under a ¯exible exchange rate, experienced a sharp contraction in its money supply in thesecond quarter of 1995, probably in an attempt to reduce the high in¯ation, which reached2,500% in December 1993.

Figure 10. Interest rate differentials, East Asia, 1997±1999



Figure 12 shows that the East Asian countries too were troubled with monetary policyduring their crisis. Indonesia is a case in point. Subsequent to the ®rst wave of the crisis, theIndonesian money supply grew rapidly, at the rate of over 25%. That was partly because ofattempts to rescue the insolvent banking sector. However, the monetary expansion did notlast very long. In the third quarter of 1998, Indonesia decreased the money supply (ÿ3%) tomeet the high interest rate policy required by the IMF.

Another case is Malaysia before the announcement of capital control on September 1998.The moderate monetary squeeze (ÿ4%) in the ®rst quarter of 1998 was probably intended to

220

215

210

25

0

5

10

15

20

25

Q11994

Q2 Q3 Q4 Q11995

Q2 Q3 Q4 Q11996

Q2 Q3 Q4

%


Figure 11. Growth rate of money supply, Latin America, 1994±1996

Figure 12. Growth rate of money supply, East Asia, 1997±1998



stabilize the currency value. In order to keep the exchange rate as steady as possible, moneybecame endogenous. More interesting still, at the onset of the crisis in December 1997, HongKong, and only Hong Kong, undertook a monetary contraction policy, in contrast to themedium monetary expansion ranging from 3% to 8% in other countries. The decrease inthe money supply was, however, so subtle (±1%) that Hong Kong was able to avoid theadjustment costs stemming from the monetary contraction necessary to preserve thecurrency board system. As far as the degree of necessary monetary contraction is considered,Hong Kong's cost seemed to be relatively small. (In terms of output loss, the story may bedifferent, as will be seen below.) The modest magnitude of the contraction can be contrastedwith the immense monetary contraction in Argentina.

2.4 Loss of output and the pace of recovery

Finally, we studied the process of recovery from the recessions following the ®nancial crises.The formal model that we introduced as a portfolio model of exchange rate determination isbased on the assumption of full employment and time-smoothing consumption assumption.The identity of nominal and real rates was assumed there. No consideration was given to theoutput loss or the interactions between exports, imports and terms of trade. Here we wouldlike to break down the process of current account recovery in the wake of the crises.

Our concern was to determine how deep the recessions had been and how long they hadlasted. The depth and the duration can be measured approximately by changes in realproduction after the crisis. We used real values of the index of industrial production (thecrude petroleum production for Indonesia and Argentina), de¯ated by the consumer priceindex. The real production index is monthly, except for Hong Kong which is quarterly. Wealso normalize the data of each country by dividing them by the level on January 1994 forLatin America (Figure 13), or by that on January 1997 for East Asia (Figure 14). (Anexception is the Brazilian IIP data, which are divided by the level of January 1995 becausethey are available only from 1995.)

Figure 13 reveals that real production in Argentina and Chile were not affected too

0.5

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

JAN1994



MAR MAY JUL SEP NOV

Jan.

199

4=1

(exc

ept B

razi

l)


Figure 13. Real production, Latin America, 1994±1996



seriously despite the Mexico crisis. At the outset of the crisis, the production indexes fellsuddenly, though temporarily, to troughs ofÿ4% in Argentina andÿ13% in Chile. The shortsubstantial decline in Chile could have been caused by a transitory increase in the countryrisk described in Section 2.2; the trough in Argentina was probably caused to a large extentby monetary contraction described in Section 2.3. We should note that the declines were notdrastic in Argentina and Chile, where only the ¯exible exchange rate system wasineffective.6

In contrast, under the ¯exible rate system, Brazil and Mexico have constantly sufferedfrom serious depressions, never recovering the production level that they had attained beforethe crisis. In these Latin American experiences, we cannot help questioning the advantageof the ¯exible exchange rate system over the other two systems.7 As a whole, the ¯exibleexchange rate system does not seem superior to the ®xed rate or the managed rate withcapital control.

Indonesia appears to be suffering a similar fate as Brazil and Mexico (Figure 14). Realproduction has decreased month by month. Of course, given the political instability inIndonesia, the decline in production cannot be attributable only to the ¯exible exchange ratesystem. In fact, South Korea has achieved the `̀ V-shaped recovery'' earlier than other Asiancountries, having begun a robust recovery after the recession of about 18 months. Similarly,Thailand will regain its original production level soon. What made the difference in thedepth and duration of recessions in Indonesia and those in other countries under the ¯exibleexchange rate system was (in addition to political turmoil) the large monetary expansion inIndonesia after the beginning of the crisis, described above. The purpose of the expansionwas to rescue the insolvent banking sector, which was considered to be a hotbed of the`̀ crony capitalism''. In this sense, the moral hazard created by the government may well

6 The only problem is that stagnation has seemed to continue until very recently in these two countries.

7 The Brazilian economy now seems to be on the rebound, showing something like the Korean style of recovery.

Figure 14. Real production, East Asia, 1997±1999



prove to have been a signi®cant aggravation of the crisis, as some authors have advocated(Corsetti et al., 1998).

Moreover, both in Hong Kong under the currency board system and in Malaysia under thecapital control, we can observe similar patterns to those countries under ®xed or managedrates in Latin America.8 Though there were moderate and temporary declines in the realproduction at the onset of the crises in Malaysia and Hong Kong, by and large the twoeconomies were thereafter navigating relatively well.9 Again, in Asia it is dif®cult to rank thealternative regimes in a linear order. One cannot claim that the IMF-led formulae werealways and necessarily better than others.

3. Closing remarks

It may be premature to give de®nitive conclusions from the short sample periods and thelimited number of countries at our disposal. Nevertheless, we can tentatively assert that nosingle regime dominates the other and that several features are worthy of attention.

(1) The domestic price levels were rather rigid at the times of shocks and during the re-sulting initial adjustments in nominal exchange rates. This highlights one favourableaspect to the overshooting exchange rate model with rigid or slow price movements.

(2) Among the countries that adopted the ¯exible rate adjustment regime after the initialshocks, Korea and Thailand have proceeded relatively smoothly and are about torebound, or have already rebounded, strongly. Indonesia had the most dif®cult ad-justment process, burdened by the country's political instability. Both Mexico and Brazilsuffered from serious recessions that lasted more than two years, and Indonesia isundergoing a similar experience.

(3) Malaysia and Chile, both countries under capital controls, seem to have coped with thedif®culties all right. As for the severity of the recovery processes, we cannot say that theIMF-led ¯oating regimes with capital mobility proved superior to those countries withcapital controls. Of course, the question remains of how to evaluate the loss of properintertemporal allocation arising from capital controls, that is, the dif®culty of in-tertemporal substitution of consumption by the public. At present, however, the effectsof the controls seem to be short-lived.

(4) As the case of Argentina indicates, the currency board system has its own problems. It isdif®cult to maintain the credibility of the currency board. The effect of monetarycontraction resulting from speculative moves was substantial. This is in contrast with thecase of Hong Kong, where the credibility issues hardly existed. Even in Hong Kong, asudden contraction of monetary policy tightened the money market, sent the short-terminterest rate into three-digit percentages, and created a substantial contraction of eco-nomic activities.

(5) The resulting adjustment costs under the ¯exible exchange rate ranged from the worstcase of still-stagnating Indonesia to the best case of fast-recovering Korea. Malaysiaunder the capital control displayed intermediate results. Hong Kong still enjoys the mostadvanced economic state in East Asia but its adjustment cost has not been negligible. As

8 The recent de-industrialization of Hong Kong may be a matter of concern.

9 Singapore is another good example, along with Hong Kong, of a very open economy adopting the ¯exiblerate. The comparison between the two countries remains to be further investigated.



for Latin America, Brazil and Mexico achieved unsatisfactory results under the ¯exibleexchange rate system, especially as documented by the slow recovery of real production.Conversely, Argentina and Chile seem to have experienced only a transitory recession inthe aftershock of the Mexican crisis.

In sum, it is hard to rank-order the performance of the three alternative regimes. Relativelymild ¯exible rates with capital control seem all right, but the costs of blocking capitalmovement may be substantial. The currency board system seems to have worked well, butthe recent stagnation of Argentina and Hong Kong casts doubt about its effectiveness.Finally, under the ¯exible rate, economies seem to have suffered a drastic downturn but theyare now recovering vigorously in terms of production. We may say that the bitter medicine ofthe IMF seems to have worked eventually for the countries receiving grants from the IMF.The approach of the IMF may be justi®ed, given the brisk recent recovery of grant-recipientcountries Thailand and Korea. At the same time, the IMF does not seem to be justi®ed inreprimanding Malaysia for blocking capital control, at least not from our statistical data.

Appendix: Data description10

(1) Nominal exchange rate: IFS series RF, the period average rates of market rates or of®cialrates, national currency units per US dollar.

(2) Real exchange rate: the real exchange rate index is derived from the nominal exchangerate, adjusted for the changes in consumer prices (IFS line 64) between the domesticcountry and the USA.

(3) Country risk: the risk premium is the residual that is calculated as the interest ratedifferential between domestic deposit rates (IFS line 60L) and the US 3-month LIBORrate (IFS line 60LDD) minus the ex post depreciation rate of the nominal exchange rate.The ex post, or realized, depreciation rate is calculated as the annual rate, that is, the rateof quarterly depreciation rate of the nominal exchange rate multiplied by 4.

(4) Money supply: the money supply growth rate is the quarterly change in M2, that is a sumof money (IFS line 34) and quasi-money (IFS line 35).

(5) Real production: the real production index is derived from the index of industrialproduction (IFS line 66) de¯ated by the consumer price, except for the crude petroleumproduction (IFS line 66AA) for Indonesia and Argentina, an exception that is oftenmade in the literature (Kaminsky and Reinhart, 1999).

Final version accepted 29 December 2000.

REFERENCES

Branson, W. H. and D. W. Henderson (1985) `̀ The Speci®cation and In¯uence of Asset Market'', in R. W. Jones andP. B. Kenen (eds.), Handbook of International Economics, Vol. 2, Amsterdam: Elsevier Science, pp. 749±805.

Corsetti, G., P. Pesenti and N. Roubini (1998) `̀ Paper Tigers? A Preliminary Assessment of the Asian Crisis'', paperprepared for NBER±Bank of Portugal International Seminar on Macroeconomics. Lisbon, 14±15 June.

Dornbusch, R. (1976) `̀ Expectations and Exchange Rate Dynamics'', Journal of Political Economy Vol. 84, No. 6,pp. 1161±76.

10 Source: International Financial Statistics (IFS), various issues.



Edwards, S. (1999) `̀ How Effective are Capital Controls?'' Journal of Economic Perspectives, Vol. 13, No. 4,pp. 65±84.

Frankel, F. A. (1999a) `̀ The International Financial Architecture'', Brookings Policy Brief no. 51, June.Ð± (1999b) `̀ No Single Currency Regime is Right for All Countries or at All Times'', Graham Lecture, Princeton

University, 20 April.Furman, J. and J. E. Stiglitz (1998) `̀ Economic Crises: Evidence and Insights from East Asia'', Brookings Papers on

Economic Activity, No. 2, pp. 1±135.Hamada, K. (2000) `̀ The Outbreak of Asian Currency Crises and their Recovery Process'', IBIC Review, no. 1,

Spring, International Bank for Economic Cooperation.International Financial Statistics (various issues) Washington, DC: International Monetary Fund.Kaminsky, G. L. and C. Reinhart (1999) `̀ The Twin Crises: The Causes of Banking and Balance-of-Payment

Problems'', American Economic Review, Vol. 89, No. 3, pp. 473±500.Kouri, P. J. K. (1976) `̀ The Exchange Rate and the Balance of Payments in the Short Run and in the Long Run'',

Scandinavian Journal of Economics, Vol. 78, pp. 280±304.Obstfeld, M. and K. Rogoff (1995) `̀ Exchange Dynamics Redux'', Journal of Political Economy, Vol. 103,

pp. 624±60.Ð± and A. M. Taylor (1997) `̀ The Great Depression as a Watershed: International Capital Mobility over the Long

Run'', in M. D. Bordo, C. Goldin and E. N. White (eds.), The De®ning Moment, The Great Depression and theAmerican Economy in the Twentieth Century, University of Chicago Press, pp. 353±402.

Radelet, S. and J. D. Sachs (1998a) `̀ The East Asian Financial Crisis: Diagnosis, Remedies, Prospects'', BrookingsPapers on Economic Activity, No. 1, pp. 1±74.

Ð± and Ð± (1998b) `̀ The Onset of the East Asian Financial Crisis'', NBER Working Paper Series no. 6680.



Documents

The Choice Between Flexible Exchange Rates, Capital Control and the Currency Board in Asian Countries: A Perspective from the “Impossible Trinity”